Prompt-Driven Low-Altitude Edge Intelligence: Modular Agents and Generative Reasoning

TL;DR

This work tackles the problem of deploying large artificial intelligence models in low-altitude edge networks by introducing P2AECF, a framework that converts high-level prompts into executable, modular reasoning workflows. It decouples cognition from fixed models through a prompt-defined cognition module, an agent-based modular execution layer, and a diffusion-based inference planner that adaptively constructs execution strategies with runtime feedback. Key contributions include a task-graph abstraction for semantic prompts, a lightweight, reusable cognitive-agent registry with dynamic mapping, and a diffusion-inspired scheduler enabling continual optimization under changing conditions; these are validated via a case study on coordinated autonomous aerial vehicle reasoning showing robust performance under latency and resource constraints. The framework promises scalable, interpretable, and self-adaptive edge cognition for mission-critical LAIN deployments, enabling real-time, context-aware collaboration between aerial and terrestrial nodes.

Abstract

The large artificial intelligence models (LAMs) show strong capabilities in perception, reasoning, and multi-modal understanding, and can enable advanced capabilities in low-altitude edge intelligence. However, the deployment of LAMs at the edge remains constrained by some fundamental limitations. First, tasks are rigidly tied to specific models, limiting the flexibility. Besides, the computational and memory demands of full-scale LAMs exceed the capacity of most edge devices. Moreover, the current inference pipelines are typically static, making it difficult to respond to real-time changes of tasks. To address these challenges, we propose a prompt-to-agent edge cognition framework (P2AECF), enabling the flexible, efficient, and adaptive edge intelligence. Specifically, P2AECF transforms high-level semantic prompts into executable reasoning workflows through three key mechanisms. First, the prompt-defined cognition parses task intent into abstract and model-agnostic representations. Second, the agent-based modular execution instantiates these tasks using lightweight and reusable cognitive agents dynamically selected based on current resource conditions. Third, the diffusion-controlled inference planning adaptively constructs and refines execution strategies by incorporating runtime feedback and system context. In addition, we illustrate the framework through a representative low-altitude intelligent network use case, showing its ability to deliver adaptive, modular, and scalable edge intelligence for real-time low-altitude aerial collaborations.

Figures (5)

• Figure 1: Representative LAIN application scenarios showing AAV-edge cooperation for collaborative inspection, aerial delivery, intelligent surveillance, disaster response, precision agriculture, and airborne data relay.
• Figure 2: The path to cognition-native edge intelligence: from LAMs to LAIN edge deployment, integrating cognitive offloading and on-device inference to enable real time, context aware operation.
• Figure 3: Challenges, unified architecture P2AECF, and applications.
• Figure 4: Prompt-driven AAV task execution, starting from prompt parsing and DAG compilation, followed by agent matching, diffusion planning, and concluding with telemetry feedback.
• Figure 5: P2AECF performance analysis.